The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill and its documentation instruct users to install the CodeRabbit CLI by piping a remote script from https://cli.coderabbit.ai/install.sh directly into the shell (evidence in SKILL.md and references/coderabbit-docs/cli.md). This pattern executes unverified remote code.
[COMMAND_EXECUTION]: The skill performs shell command execution including gh pr checkout, git commit, and coderabbit review to interact with the local environment and git provider as described in SKILL.md and cli-skills.md.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to download installation scripts and interact with the CodeRabbit API at api.coderabbit.ai for audit logs and metrics (evidence in api-reference-audit-logs.md).
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection vulnerability surface. 1. Ingestion points: The skill reads untrusted data from pull request review threads, comments, and issue tracker descriptions (file: SKILL.md). 2. Boundary markers: Absent. The agent is not provided with explicit delimiters or instructions to ignore embedded commands. 3. Capability inventory: The skill can execute arbitrary CLI tools, git commands, and network operations (file: cli-skills.md). 4. Sanitization: Absent. External content from PR threads is used directly to inform automated code fixes without evidence of validation.

coderabbit